Anti-VEGF treatment reduces blood supply andincreases tumor cell invasion in glioblastomaOlivier Keunena,b, Mikael Johanssona,c, Anaïs Oudina, Morgane Sanzeya, Siti A. Abdul Rahima, Fred Facka,Frits Thorsenb, Torfinn Taxtb,d, Michal Bartose, Radovan Jirike,f, Hrvoje Mileticb,g, Jian Wangb, Daniel Stiebera,Linda Stuhrb, Ingrid Moenb, Cecilie Brekke Ryghb, Rolf Bjerkviga,b,1, and Simone P. Nicloua,1,2

aNorLux Neuro-Oncology Laboratory, Oncology Department, Centre de Recherche Public de la Santé, 1526 Luxembourg, Luxembourg; bDepartment ofBiomedicine, University of Bergen, 5009 Bergen, Norway; cDepartment of Radiation Sciences, Oncology, Umeå University, 90185 Umeå, Sweden; dDepartmentof Radiology, Haukeland University Hospital, 5021 Bergen, Norway; eInstitute of Scientific Instruments, Academy of Sciences of the Czech Republic, 61264Brno, Czech Republic; fDepartment of Biomedical Engineering, Brno University of Technology, 61200 Brno, Czech Republic; and gDepartment of Pathology,Haukeland University Hospital, Bergen, Norway

Edited* by George Klein, Karolinska Institute, Stockholm, Sweden, and approved January 21, 2011 (received for review October 6, 2010)

Bevacizumab, an antibody against vascular endothelial growthfactor (VEGF), is a promising, yet controversial, drug in humanglioblastoma treatment (GBM). Its effects on tumor burden, re-currence, and vascular physiology are unclear. We therefore de-termined the tumor response to bevacizumab at the phenotypic,physiological, and molecular level in a clinically relevant intracranialGBM xenograft model derived from patient tumor spheroids. Usinganatomical and physiological magnetic resonance imaging (MRI),we show that bevacizumab causes a strong decrease in contrastenhancementwhile having only amarginal effect on tumor growth.Interestingly, dynamic contrast-enhancedMRI revealed a significantreduction of the vascular supply, as evidenced by a decrease in in-tratumoral blood flow and volume and, at the morphological level,by a strong reduction of large- and medium-sized blood vessels.Electron microscopy revealed fewer mitochondria in the treatedtumor cells. Importantly, this was accompanied by a 68% increase ininfiltrating tumor cells in the brain parenchyma. At the molecularlevel we observed an increase in lactate and alanine metabolites,together with an induction of hypoxia-inducible factor 1α andan activation of the phosphatidyl-inositol-3-kinase pathway. Thesedata strongly suggest that vascular remodeling induced by anti-VEGF treatment leads to a more hypoxic tumor microenvironment.This favors a metabolic change in the tumor cells toward glycolysis,which leads to enhanced tumor cell invasion into the normal brain.The present work underlines the need to combine anti-angiogenictreatment in GBMs with drugs targeting specific signaling or meta-bolic pathways linked to the glycolytic phenotype.

angiogenesis | glioma | metabolism | perfusion

Glioblastomas (GBMs) are highly vascularized brain tumorsand are therefore attractive targets for anti-angiogenic ther-

apies (1). In particular, vascular endothelial growth factor(VEGF) has been identified as a critical regulator of angiogenesis,and currently a number of clinical trials targeting the VEGF-signaling pathways are under development (2, 3). Bevacizumab(bev), a humanized anti-VEGF antibody, has shown promisingresults in exploratory phase II trials of recurrent GBM.Alone or incombination with irinotecan, it is well tolerated and shows a highradiological response rate and possibly an increase in medianprogression-free survival compared with historical controls (4–7),although no impact on overall survival has been reported (8).However, these results are based on small patient cohorts and,because anti-angiogenic agents directly affect vessel permeability,the imaging response assessment based on contrast enhancement(CE) is highly ambiguous (9). Indeed, a direct antitumor effect ofbev has remained elusive and the infiltrative part of the tumormay even increase (10, 11). In addition to a lack of robust clinicaldata, the cellular and molecular consequences of anti-VEGFtreatment have not been outlined (12). Detailed information onhow bev affects GBM is important not only to understanding thesuccess or failure of such treatment, but also to providing educatedadvice on how combination therapies should be optimized.

We have developed a clinically highly relevant human GBMmodel in rats that fully reflects the growth pattern of humantumors in situ, including extensive infiltration into the brain pa-renchyma (major pattern of invasion along fiber tracts and bloodvessels), prominent angiogenesis, and necrosis (13–15). Compar-ative genomic hybridization studies have confirmed the geneticsimilarity between xenografted tumors and the correspondinghuman tumors (Fig. S1). Thus, our model retains the cellular andgenetic heterogeneity that characterizes human GBMs. This is ofprime importance for experimental studies, because it is wellknown that xenograft models based on glioma cell lines poorlyreflect the clinical situation. Using our patient-based model sys-tem, we have assessed the phenotypic, physiological, and molec-ular effects of bev treatment in GBMs. We show that anti-angiogenic treatment leads to major vessel remodeling, resultingin reduced perfusion and an increase in hypoxia in the tumormicroenvironment. This leads to a metabolic shift in the tumorstoward glycolysis, reflected by both an increase in lactate pro-duction and a stabilization of hypoxia-inducible factor 1α (HIF1α),and is accompanied by a dramatic increase in cell invasion intothe normal brain.

ResultsMarked Decrease in Contrast Agent Leakage After BevacizumabTreatment and a Limited Reduction in Tumor Progression. Threeweeks after implantation of patient-derived tumor spheroids,animals were randomly divided into control groups and treat-ment groups. Bevacizumab was administered through weekly i.v.injections over a 3-wk period. Initial tumor size before treatmentwas 23.1 ± 9.6 mm3 (values are reported as mean ± SD, unlessotherwise specified) as determined from the visible part (tumorcore) of T2-weighted MRI images. After 3 wk of treatment, thetumors in the control group had grown to an average size of322 ± 176 mm3, whereas the bev-treated tumors had reached anaverage size of 230 ± 76 mm3 (see Fig. 1 A and B for repre-sentative T2-weighted images). The tumor doubling time (TDT)(16) of bev-treated tumors increased by about 16% (P < 0.05)compared with untreated tumors, indicating a slowdown of tumorprogression during treatment (Fig. 1C). Comparable results wereobtained with immunohistochemical staining for the proliferation

Author contributions: O.K., M.J., R.B., and S.P.N. designed research; O.K., M.J., A.O., M.S.,S.A.A.R., F.F., F.T., H.M., J.W., D.S., C.B.R., L.S., I.M., and R.B. performed research; T.T.,M.B., and R.J. contributed new reagents/analytic tools; O.K., M.J., A.O., M.S., S.A.A.R., F.F.,F.T., T.T., M.B., R.J., H.M., D.S., R.B., and S.P.N. analyzed data; and O.K., R.B., and S.P.N.wrote the paper.

The authors declare no conflict of interest.

*This Direct Submission article had a prearranged editor.

Freely available online through the PNAS open access option.1R.B. and S.P.N. contributed equally to this article.2To whom correspondence should be addressed. E-mail: simone.niclou@crp-sante.lu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1014480108/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1014480108 PNAS | March 1, 2011 | vol. 108 | no. 9 | 3749–3754

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

Mar

ch 2

, 202

1

marker Ki67, indicating a small but significant reduction ofstained nuclei (11% reduction, P < 0.05) in the treated animals(Fig. 1 D–F). Average tumor apparent diffusion coefficient(ADC) values in the tumor core were statistically unchanged,suggesting that there was no major effect on the cellularity of thetumor. Tumor volumes assessed from postcontrast T1-weightedsequences provided similar values as those assessed from T2-weighted sequences (Fig. 1 G–I), indicating a reliable estimationof tumor volume independent of CE and suggesting a limitedamount of edema in our model. As expected, T1-weighted imagesobtained after injection of contrast agent and quantification ofthe area under the curve (AUC) from the dynamic T1-weightedimages revealed a significant reduction (34%, P < 0.05) in CE inthe bev-treated group (Fig. 1 J–L). This is in full agreement withreported clinical data (4, 17, 18). In summary, these data indicatethat, in our xenograft model derived from patient tumor material,bev induces a slight reduction in tumor progression during thetreatment period and strongly reduces leakage of contrast agentfrom the blood vessels.

Bevacizumab Reduces Tumor Blood Flow and Blood Volume andAffects Vessel Permeability Parameters. To address the physiolog-ical consequences of bev on the tumor vasculature and gain in-sight into the CE changes observed, we performed dynamiccontrast-enhanced MRI (DCE-MRI). In this technique, thetemporal changes in contrast consecutive to the bolus injection ofthe contrast agent are recorded and fed to a pharmacokineticmulticompartmental model from which perfusion and perme-

ability parameters are estimated. Several existing models differ inthe number of available parameters, complexity, and stability.Thus, the Tofts model (19), which is often used in clinical settings,provides access to three independent parameters [blood orplasma volume (vb), extravascular extracellular space fraction(ve), and blood-to-tissue transfer constant (Ktrans)]. The tissuehomogeneity model (20) used in this study provides the additionalparameter of blood flow. Interestingly, we observed in bev-treatedanimals a 17% reduction in tumor blood flow (Fb, P< 0.01) (Fig. 2A–C) and a 46% reduction in blood volume per unit of tissue (vb,P < 0.001) (Fig. 2 D–F), suggesting a reduced supply of oxygenand nutrients to the tumor. As expected from the reduced CEseen on radiological images (Fig. 1 J–L), several permeabilityparameters were significantly reduced by the treatment: Ktrans(33% reduction, P < 0.001) (Fig. 2 G–I), permeability surface(PS), extraction fraction (E), and the tissue-to-blood backflowconstant (kep). No significant changes were detected in the vefraction (Fig. 2 J–L).Using statistical analysis of variance, we determined the origin

of the variability of the perfusion parameters: although tumorsize and animal weight contributed to the variation, the treat-ment itself had a distinct statistically significant effect on thereduction of blood flow, blood volume, transit time, permeabilitysurface area, and Ktrans. Similar trends for vb, Ktrans, kep, and vewere also observed when the Tofts model was used instead of thetissue homogeneity model, thus establishing that the changes inperfusion parameters resulted from the treatment effect in-dependent of the model used. Taken together, these data dem-

Fig. 1. Quantification of tumor progression and contrast enhancement.Representative images of control (A, D, G, and J) and bev-treated animals (B,E, H, and K). Tumor volume was assessed from T2-weighted images (A and B)to determine tumor doubling time (TDT) (C), Ki67 immunostaining (D andE), and quantification thereof (F). Control values were set at 100%. (G andH) Postcontrast T1-weighted images. (I) Tumor volumes assessed frompostcontrast T1-weighted images were similar to T2-weighted images. (J–L)Reduction of contrast agent uptake (CE) in the treated group as evidencedby the mean tumor area under the curve (AUC). Ctrl: controls; Tr: treated.(Scale bars: ±SE.) *P < 0.05.

Fig. 2. DCE-MRI analysis of bev-treated glioblastomas. Tumor perfusionmaps of representative control (A, D, G, and J) and bev-treated animals (B, E,H, K). Bevacizumab led to a significant reduction of blood flow, Fb (A–C); ofblood volume per unit of tissue, vb (D–F); and of the blood-to-tissue ex-traction constant, Ktrans (G–I). (J–L) The extravascular extracellular space(interstitial space volume) per unit of tissue, ve, was not significantly modi-fied. Ctrl: controls; Tr: treated. (Scale bars: ± SE.) *P < 0.05, **P < 0.01, ***P <0.001. Colors range from blue (low values) to red (high values).

3750 | www.pnas.org/cgi/doi/10.1073/pnas.1014480108 Keunen et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

, 202

1

onstrate that the loss of CE observed after anti-VEGF treatmentresults not only from reduced vessel permeability but also froma reduction of blood flow. This was rather unexpected as it hadbeen proposed that antivascular therapy leads to a “normaliza-tion” of the tumor vasculature at the morphological level, whichwas thought to be accompanied by increased perfusion and im-proved tumor blood supply (21). Our data show that, at least inGBM, anti-angiogenic treatment leads to reduced blood supplyand consequently reduced oxygenation of the tumor bed. Ittherefore appears that a “morphological” normalization of thevascular bed is not necessarily accompanied by a “functional”normalization of the vascular supply.

Anti-VEGF Treatment Reduces Vessel Density and Strongly IncreasesCell Invasion. Quantification of endothelial cell staining revealeda severe reduction in the number of vessels in the bev-treatedanimals (Fig. 3 A–C). In particular, large-sized (58%, P < 0.001)and medium-sized (17%, P < 0.001) blood vessels were stronglyreduced, indicating a strong normalizing effect of the treatment.The number of small-sized vessels with a more regular appear-ance was not affected (Fig. 3C).To visualize human tumor cells in the rat brain and to de-

termine their distribution outside the tumor core, we used a hu-man-specific nestin staining, a protein expressed by the vastmajority of GBM cells (15, 22, 23). Interestingly, treated tumorsrevealed a more homogeneous morphology compared with thecontrols (Fig. 3 D–F). For example, control tumors (Fig. 3D)showed abundant areas of necrosis and large blood vessels, whichwere virtually absent in the treated samples (Fig. 3E). Increasedtumor homogeneity in the treated samples was also confirmed byT2-weighted MRI images (Fig. 3F). Importantly, a strong andhighly significant increase in the number of tumor cells invadingthe normal brain (68%, P < 0.001) was measured in the bev-treated group (Fig. 3 G–I). In addition to the number of invadingcells, the distance of infiltrating cells from the tumor core wasstrikingly higher in bev-treated xenografts. The switch to a moreinvasive phenotype also correlated with a decrease in apparentdiffusion coefficient (ADC) in the tumor periphery of the treated

animals, although statistical significance was not reached in thisparameter. In summary, we show that bev induces a reduction oflarge- and medium-sized blood vessels, decreases tumor het-erogeneity, and results in a dramatic increase in parenchymaltumor cell infiltration.

Histological and Ultrastructural Changes Induced by BevacizumabTreatment.The untreated xenografts displayed typical hallmarks ofGBM, indicated by pseudopalisading necrotic areas and micro-vascular proliferations in the tumor center and in the periphery(Fig. 4 A and C). In comparison, treated tumors were mostly de-void of these features, the vessels displayed a nonproliferatingendothelium, and the cell density in the tumor periphery seemedreduced (Fig. 4 B and D). Transmission electron microscopyrevealed, in the untreated tumors, a strong activity of the vascularendothelium with sprouting processes (Fig. 4E). In comparison,the blood vessels in the treated tumors were more normalizedwith less visible sprouts (Fig. 4F). However, no apparent signs ofastrocytic end-feet were seen in these vessels, indicating that thereconstitution of the blood–brain barrier may not be complete. Inthe tumor core, the treated tumors showedmicro-areas of cell lysisindicative of cell death (Fig. 4 G and H). Quantification of thenumber of mitochondria revealed a strong reduction of mito-chondria in the tumor cells after bev treatment (Fig. 4I). In-terestingly, ultrastructural analysis of the invasive front, in contrastto the tumor core, showed more homogeneous, rather looselyconnected tumor cells (Fig. 4J). In summary, we found a loss ofendothelial cell proliferation in bev-treated tumors and a reducednumber of mitochondria in the tumor cells, which may be a con-sequence of reduced tumor oxygenation.

Increased Tumor Hypoxia and Activation of the PI3K- and Wnt-Signaling Pathways After Anti-VEGF Treatment. We have pre-viously shown that the infiltrative cells within GBMs show atendency toward anaerobic glycolysis indicated by increasedlactate production (24). In bev-treated tumors, MR spectroscopyshowed a tendency toward an accumulation of lactate, alanine,choline, myo-inositol, creatine, taurine, and mobile lipids (TableS1), a combination that has previously been associated with in-creased hypoxia in human brain tumor spectra (25). Increasedlactate levels were also confirmed by proton NMR. Western blotanalysis revealed an increase in HIF1α protein in the treatmentgroup compared with the untreated samples (Fig. 5 A and B),which was also confirmed at the mRNA level (Table S3).Gene expression analysis of the xenografts further revealed,

after treatment, an up-regulation of gene transcripts involved inthe phosphoinositol-3-kinase (PI3K)- and Wnt-signaling path-ways (Table S2). In fact, from 84 genes related to the PI3K/Aktpathway, 44 were up-regulated more than 1.5-fold after bevtreatment, and only one gene was down-regulated in the sameorder of magnitude. Although not all genes were statisticallysignificant (Table S2) and some negative regulators of the path-way were induced (PTEN, TSC1/2), overall there seems to bea clear trend toward activation of the pathway (Fig. 5C). In-terestingly, a similar trend was seen for the Wnt pathway,which we have previously found to be induced in the infiltrativecompartment of GBMs (13). PCR analysis also indicated that,as a result of bev treatment, the tumor up-regulates severalangiogenesis-related transcripts, suggesting an induction of al-ternative angiogenic pathways (Table S3). These include an-giopoietin 2, prostaglandin-endoperoxide synthase 1, urokinase,endothelial tyrosine kinase, and VEGF-A (Table S3). In sum-mary, we show that, in GBM xenografts, bev increases tumorhypoxia and activates alternative angiogenic pathways and mo-lecular functions associated with stem cell biology and the in-vasive phenotype.

DiscussionGlioblastomas are highly vascularized tumors and therefore rep-resent attractive targets for anti-angiogenic therapies. Despiteimpressive radiological responses on CET1-weightedMRI images

Fig. 3. Changes in blood vessel morphology and tumor cell invasion afterbev treatment. Immunostaining for von Willebrand factor (vWF) (A and B)and quantification thereof (C), indicating a significant reduction in thedensity of medium and large blood vessels and in total vessel number afterbev treatment. (Scale bar: 200 μm.) Nestin-stained composite images (D andE) reveal a more homogeneous appearance of the treated compared withuntreated tumors, also reflected in corresponding T2-weighted MRI images(F). Large vessels (“V”) appear as dark tortuous lines in nestin and T2-weighted images and necrotic areas (“N”) as brighter spots. Quantificationof the nestin-positive cells outside the tumor core (G and H) shows a 68%increase in cell invasion after treatment (I). mi.v: microvessels; in.v:intermediate-sized vessels; ma.v: macrovessels; Ctrl: controls; Tr: treated.(Scale bars: ± SE.) ***P < 0.001.

Keunen et al. PNAS | March 1, 2011 | vol. 108 | no. 9 | 3751

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

Mar

ch 2

, 202

1

in bev-treated patients, tumor cell invasion and recurrence remainmajor challenges. Thus, there is a strong need to improve treat-ment strategies for glioblastoma and to better understand themechanisms of failure for the targeted anti-angiogenic therapies.Here we address these mechanisms by using clinically highlyrelevant GBM xenografts. Our data show that anti-VEGF treat-ment induces the loss of large-sized vessels, a dramatic increase intumor cell invasion, and a significant reduction in tumor bloodflow and blood volume. An in-depth quantification of the physi-ology of the tumor vasculature further revealed a reduction in

several permeability parameters, including Ktrans, kep, the perme-ability surface area PS, and the extraction fraction E. Increasedtumor hypoxia after anti-VEGF treatment is suggested by in-duction of HIF1α (both at the gene and the protein level) and anincrease in glycolytic metabolites (e.g., lactate). This was accom-panied by an up-regulation of the PI3K- and Wnt-signaling path-ways. Interestingly, we also detected up-regulation of severalangiogenesis-related genes, including VEGF-A. However, the li-gand is likely to be efficiently cleared by the excess of drug, andadditional compensatory angiogenic factors, althoughmaintainingexisting endothelial cells, appear not sufficient to induce neo-angiogenesis and to increase blood flow. It should be noted thatthe xenografts used in the present study carry a typical geneticGBM signature such as amplification of chromosome 7 includingEGFR; deletion of CDKN2A/B and loss of one copy of chromo-some 9; hemizygous deletion of chromosome 10 including thePTEN gene; and deletion of PI3KR1 (Fig. S1). It is thereforehighly unlikely that the observed effects of anti-angiogenic treat-ment are unique to this particular tumor; however, it cannot beruled out that certain GBMs respond differently to bev treatment.In view of recent findings indicating that GBMs can contribute totheir own vascular supply (26, 27), it will be of interest to de-termine the VEGF dependency of these tumor-derived endothe-lial cells and their response to anti-angiogenic treatment.Bevacizumab is already in clinical use for breast, lung, and

colon cancer and appears also promising for the treatment ofGBMs. There is, however, a controversy regarding treatmentefficacy in terms of patient survival and the validity of radio-logical response rates as a surrogate endpoint for clinical benefit(9, 27, 28). Furthermore, it is currently not clear whether ra-diotherapy, which is dependent on the oxygenation level of thetumor, and/or systemic drug delivery, which is influenced byvessel permeability and blood flow, should benefit from anti-angiogenic treatment or not (17). To address these questions,a better understanding of the biological effects of anti-VEGFtreatment is mandatory. Detailed analysis of DCE-MRI param-eters as performed here was possible due to the use of an ad-vanced pharmacokinetic tissue homogeneity model (20), which,in comparison with the Tofts model traditionally used in theclinical setting (19, 29), provides access to the additional impor-tant parameter of blood flow. Because increasing the number offree variables can possibly impact the stability of the model, dif-ferent measures were taken to optimize the model, among whichwere the adoption of a good time resolution, the use of localarterial input functions, and a careful monitoring of the model fit(see SI Materials and Methods for details). The reduction of CE inT1-weighted images observed after anti-angiogenic treatment cantherefore be attributed to a combined effect of a reduction inblood volume, blood flow, and vessel permeability. In agreementwith previous reports (30, 31), we propose that static T1-weightedsequences with CE alone are not sufficient to properly assess theefficacy of anti-angiogenic therapies and that perfusion parame-ters obtained from DCE sequences provide useful additional in-sight. It should, however, be noted that current models in use havelimitations and different applicabilities. Selecting a basic modelmay provide advantages in terms of simplicity of implementation,but may fail to properly assess the physiological changes inducedby the treatment. Diffusion-weighted imaging may have an ad-ditional role in assessing changes in cellularity induced by thetreatment (32, 33), provided that the causative factors (invasivecells, necrosis, edema) can be distinguished.Early work from the pioneers of anti-angiogenic treatment

suggested that the reduction of newly formed blood vesselsstarves the tumor from nutrients and oxygen, thereby reducingtumor growth and inducing tumor cell death (starvation hypoth-esis) (34). It has, however, become clear that the mechanisms ofanti-angiogenic therapy are more complex and may also dependon the tumor type (35). For GBMs, at least, induction of tumorcell death has not been demonstrated. More recently, it wasproposed that anti-angiogenic treatment leads to blood vesselnormalization, thought to be accompanied by increased blood

Fig. 4. Histological and ultrastructural changes after bev treatment.Hematoxylin- and eosin-stained sections of GBM xenografts (A–D). In controltumors (A and C), typical hallmarks of GBM growth are visible: necrosis (“N”)and microvascular proliferations (arrowheads) in the tumor center (A) andperiphery (C), but not in treated tumors (B and D). The endothelium appearsmore normal in treated tumors (arrows). [Scale bar (A–D): 50 μm.] (E–H and J)Transmission electron microscopy (TEM) images of GBM xenografts. Micro-vascular proliferation and endothelial cell sprouting (red arrows) in controltumors (E). Treated tumors show more normalized blood vessels, yet nomature blood–brain barrier (F). A denser cellular composition in the tumorcore in control tumors (G) compared with the treatment group (H) whereseveral lytic areas were observed (red arrow). Cells from the invasive front (J)had a more elongated morphology, suggesting a subpopulation within thetumor. Dividing cells in the invasive front (white arrow in J). [Scale bar (E–Hand J): 5 μm.] Quantification of mitochondria per cell from TEM micrographs(I). Ctrl: controls; Tr: treated. (Scale bars: ±SE.) **P < 0.01.

3752 | www.pnas.org/cgi/doi/10.1073/pnas.1014480108 Keunen et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

, 202

1

flow and oxygenation (normalization hypothesis) (1, 21, 36). Incontrast, our data demonstrate a reduction in tumor perfusionand oxygenation in the GBM xenografts. Although increasedblood flowmight occur during a short normalization window (36),our data suggest that, in GBMs, the long-term effects of anti-VEGF agents are increased hypoxia and increased invasive po-tential. An anaerobic metabolism is also reflected by an elevationof metabolites associated with glycolysis (Table S1) and the in-duction of HIF1α protein (Fig. 5). These are phenomena thatoften correlate with increased invasion and metastasis in solidtumors (37). In this context, it is also interesting to note that wefound a significant reduction of mitochondria per cell in bev-treated tumors (Fig. 4I).Enhanced cell infiltration after anti-angiogenic treatment has

also been reported in other tumor models (38, 39). Our resultsfurther show an up-regulation of the PI3K- and the Wnt-signaling pathways as a result of bev treatment (Fig. 5C). Thesetwo signaling networks earlier were linked to invasive cells withinGBMs (13, 40, 41). Interestingly, the PI3K/Akt pathway is alsocritically involved in several steps of both anaerobic and aerobicglycolysis regulation, as, for example, in localization of glucosetransporters at the cell surface and maintenance of hexokinasefunction in the absence of extrinsic regulatory factors (40, 42,43). Thus, activation of Akt as a result of bev-induced hypoxiamay increase intracellular glucose and stimulate anaerobic gly-colysis and lactate production, thereby promoting invasiveness.In conclusion, a VEGF blockade causes only a small reduction

in tumor burden, but does induce a strong depletion of large-and intermediate-sized blood vessels with a subsequent re-duction in vascular leakage and intratumoral blood flow. Anti-VEGF treatment strongly increases tumor cell invasion, whichmay result from increased hypoxia in the tumor microenviron-ment. These data are of major clinical importance with regardto combination therapies. For example, our data suggest thatradiotherapy, partly dependent on the oxygenation level of thetumor, and systemic drug delivery, influenced by vessel perme-ability and blood flow, may not profit from coadministered anti-

angiogenic treatment in GBM. Bevacizumab is currently ap-proved by the Food and Drug Administration for second-linetreatment of GBMs, and new clinical trials aim at assessing itspotential in first-line treatment, possibly with additional che-motherapeutic compounds. Our data suggest possible metabolicadaptation mechanisms that might compromise the success ofsuch trials. We propose that anti-angiogenic therapy could ben-efit from the adjuvent delivery of drugs targeting the HIF1α andPI3K/Akt pathways or by directly interfering with the glycolyticmetabolism of tumor cells. Such drugs have recently shownpromising results for the treatment of malignant glioma both inexperimental models and in patients (44–47).

Materials and MethodsTumor Material. Patient GBM-derived spheroids (from patient P3) wereexpanded through serial transplantation in nude rats, thus generatinga standardized pool of spheroids (300–400 μm) and giving rise to phenotyp-ically identical (highly invasive and highly angiogenic) GBMs in all xenografts(14). The same genetic aberrations were present in the primary biopsy and inresulting xenografts as determined by array comparative genomic hybrid-ization (Fig. S1). Collection of human biopsy tissue was approved by the re-gional ethical committee (Haukeland University Hospital, Bergen, Norway).

Intracranial Implantation and Treatment. Ten GBM spheroids were stereo-tactically implanted into the brain (posterior to the bregma and 3 mm to theright of the midline suture at a depth of 2.5 mm) of 30 athymic nude rats(rnu−/rnu−) as described (14). Surgical procedures were in accordance withthe Norwegian Animal Act and the local ethical committee. Tumor take wasverified by MRI 3 wk post-implantation, and animals were stratified intocontrol (n = 15) and treatment groups (n = 15). Bevacizumab (10 mg/kg) wasgiven weekly by i.v. injections into the tail vein in accordance with currentclinical practice. Control animals were not injected. After 3 wk, rats un-derwent MRI imaging after which they were killed by perfusion fixation (4%paraformaldehyde/PBS) or by decapitation followed by tumor dissection forRNA isolation, protein extraction, and electron microscopy. No treatment-related adverse effects were observed during the study.

Fig. 5. Molecular changes induced in GBM xenografts after bevtreatment. Western blot for HIF1α in control and bev-treatedglioblastomaxenografts (A) andquantification thereof (B). Signalnormalization with a human-specific nestin antibody (n = 7). (C)Schematic of key regulatory molecules associated with receptortyrosine kinase activation induced after bev treatment (genesmarked in dark pink were up-regulated >1.5-fold; genes in lightpink were up-regulated >1-fold ; genes in gray were unchanged;and genes in white were not on the array). (Scale bars: ±SE.) *P <0.05. (See also Tables S2and S3.) Ctrl: controls; Tr: treated. AKT1/2:protein kinase B/Bβ; APC: adenomatous polyposis coli; AXIN: axisinhibition protein; βcat: β1 Catenin (CTNNB1); EGFR: epidermalgrowth factor receptor; Erk: mitogen-activated protein kinase 1(MAPK1); FOX01: forkheadboxO1; FZD: Frizzled homolog; GLUT:glucose transporter; Grb2: growth factor receptor-bound protein2; GSK3B: glycogen synthase kinase 3 β; IGFR: insuline-like growthfactor receptor; LEF1: lymphoid enhancer-binding factor 1;mTOR: FK506 binding protein 12-rapamycin associated protein 1;PDGFR: platelet-derived growth factor receptor; PI3K: phosphoi-nositide-3-kinase; PTEN: phosphatase and tensin homolog; RAS:RAS protein superfamily; p70s6k: ribosomal protein S6 kinase;SHC: (Src homology 2 domain containing) transformingprotein 1;Sos: Son of sevenless; TCF3: transcription factor 3; TSC1/2: tuber-ous sclerosis 1/2.

Keunen et al. PNAS | March 1, 2011 | vol. 108 | no. 9 | 3753

MED

ICALSC

IENCE

S

Dow

nloa

ded

by g

uest

on

Mar

ch 2

, 202

1

MRI and MRS. MR images were acquired using a 7T Pharmascan small-animalMR scanner (Bruker Biospin) with a linear rat head transmitter/receiver coil.The animals were anesthetized with 1–2% isofluranemixed with 50% air and50% O2 and placed in a prone position in a cradle containing a heating padat 37 °C. Breathing was monitored throughout. MR sequences used includeT2- and T1-weighted images before and after injection of the contrast agent(Gadodiamide, Omniscan; GE Healthcare), diffusion weighted imaging, high-speed DCE-MRI (FLASH sequence with a time resolution of 1.5 s), and 1HMRS. Tumor volumes, TDT, T2/T1 volume ratio (indicative of edema extend),and CE were calculated in nordicICE (NordicNeuroLab) after delineatingthe tumor on consecutive sections. ADCs were calculated in Paravision 5(Bruker Biospin). Perfusion and permeability parameters were calculated onthe basis of the tissue homogeneity model (20) by using routines customdeveloped in Matlab (MathWorks). Tumor masks were applied on twoconsecutive slices of the parameter maps obtained to provide average per-fusion parameters. The Tofts model was used for comparison (nordicICE2.3.4; NordicNeuroLab). Spectra obtained from MRS were processed inLCModel (48) to obtain absolute metabolite concentrations and reported asmedian values for the treated and control animal groups. For detailed in-formation, see SI Materials and Methods.

Immunohistochemistry. Coronal paraffin sections (7–10 μm) were used forhistology (hematoxylin/eosin staining) and immunohistochemical analysis:human-specific antibody against nestin (MAB5326; Millipore 1:200), von Wil-lebrand factor (A0082; Dako; 1:200), and Ki67 (M7240; Dako, 1:75). Stainingwas performed according to the manufacturer’s instructions (Envision kitK4011/K4007; Dako). Angiogenesis, invasiveness, and proliferation rate weredetermined as follows. Stained blood vessels were arbitrarily divided intothree classes according to their size. Vessel count indices were computed asa percentage of vessels in treated animals versus controls. Proliferation wasassessed by counting KI67-positive cells per section. Invasive potential wasassessed by counting nestin-positive invasive cells around the tumor core. Forall other experimental procedures, see SI Materials and Methods.

ACKNOWLEDGMENTS. We thank A. Muller and P. Nazarov for advice oningenuity pathway analysis and statistics, respectively, and T. Pavlin andK. Brandt for their assistance in MRI troubleshooting. The project was sup-ported by Centre de Recherche Public de la Santé through a grant from theMinistry of Research and Higher Education in Luxembourg, the Fonds Na-tional de la Recherche of Luxembourg; by the Norwegian Cancer Society, theNorwegian Research Council, Innovest AS, Helse Vest, Haukeland UniversityHospital, and the Bergen Medical Research Fund; and by the Czech ScienceFoundation (Project Grant GA102/09/1690).

1. Jain RK, et al. (2007) Angiogenesis in brain tumours. Nat Rev Neurosci 8:610–622.2. Reardon DA, Wen PY, Desjardins A, Batchelor TT, Vredenburgh JJ (2008) Glioblastoma

multiforme: An emerging paradigm of anti-VEGF therapy. Expert Opin Biol Ther 8:541–553.

3. Miletic H, Niclou SP, Johansson M, Bjerkvig R (2009) Anti-VEGF therapies formalignant glioma: Treatment effects and escape mechanisms. Expert Opin TherTargets 13:455–468.

4. Vredenburgh JJ, et al. (2007) Phase II trial of bevacizumab and irinotecan in recurrentmalignant glioma. Clin Cancer Res 13:1253–1259.

5. Friedman HS, et al. (2009) Bevacizumab alone and in combination with irinotecan inrecurrent glioblastoma. J Clin Oncol 27:4733–4740.

6. Vredenburgh JJ, et al. (2007) Bevacizumab plus irinotecan in recurrent glioblastomamultiforme. J Clin Oncol 25:4722–4729.

7. Kreisl TN, et al. (2009) Phase II trial of single-agent bevacizumab followed bybevacizumab plus irinotecan at tumor progression in recurrent glioblastoma. J ClinOncol 27:740–745.

8. Norden AD, et al. (2009) An exploratory survival analysis of anti-angiogenic therapyfor recurrent malignant glioma. J Neurooncol 92:149–155.

9. van den Bent MJ, Vogelbaum MA, Wen PY, Macdonald DR, Chang SM (2009) Endpoint assessment in gliomas: Novel treatments limit usefulness of classicalMacdonald’s criteria. J Clin Oncol 27:2905–2908.

10. Norden AD, et al. (2008) Bevacizumab for recurrent malignant gliomas: Efficacy,toxicity, and patterns of recurrence. Neurology 70:779–787.

11. Verhoeff JJ, et al. (2009) Concerns about anti-angiogenic treatment in patients withglioblastoma multiforme. BMC Cancer 9:444.

12. Ellis LM, Hicklin DJ (2008) VEGF-targeted therapy: Mechanisms of anti-tumouractivity. Nat Rev Cancer 8:579–591.

13. Sakariassen PO, et al. (2006) Angiogenesis-independent tumor growth mediated bystem-like cancer cells. Proc Natl Acad Sci USA 103:16466–16471.

14. Wang J, et al. (2009) A reproducible brain tumour model established from humanglioblastoma biopsies. BMC Cancer 9:465.

15. Rajcevic U, et al. (2009) iTRAQ-based proteomics profiling reveals increased metabolicactivity and cellular cross-talk in angiogenic compared with invasive glioblastomaphenotype. Mol Cell Proteomics 8:2595–2612.

16. Nakajima M, Nakasu S, Morikawa S, Inubushi T (1998) Estimation of volume doublingtime and cell loss in an experimental rat glioma model in vivo. Acta Neurochir (Wien)140(6):607–612; discussion 612–603.

17. Norden AD, Drappatz J, Wen PY (2009) Antiangiogenic therapies for high-gradeglioma. Nat Rev Neurol 5:610–620.

18. Pope WB, Lai A, Nghiemphu P, Mischel P, Cloughesy TF (2006) MRI in patients withhigh-grade gliomas treated with bevacizumab and chemotherapy. Neurology 66:1258–1260.

19. Tofts PS, et al. (1999) Estimating kinetic parameters from dynamic contrast-enhancedT(1)-weighted MRI of a diffusable tracer: Standardized quantities and symbols.J Magn Reson Imaging 10:223–232.

20. Koh TS, et al. (2001) The inclusion of capillary distribution in the adiabatic tissuehomogeneity model of blood flow. Phys Med Biol 46:1519–1538.

21. Jain RK (2005) Normalization of tumor vasculature: An emerging concept inantiangiogenic therapy. Science 307:58–62.

22. Colin C, et al. (2007) Relevance of combinatorial profiles of intermediate filamentsand transcription factors for glioma histogenesis. Neuropathol Appl Neurobiol 33:431–439.

23. Kitai R, et al. (2010) Nestin expression in astrocytic tumors delineates tumorinfiltration. Brain Tumor Pathol 27:17–21.

24. Thorsen F, et al. (2008) Two distinct tumor phenotypes isolated from glioblastomasshow different MRS characteristics. NMR Biomed 21:830–838.

25. Howe FA, et al. (2003) Metabolic profiles of human brain tumors using quantitative invivo 1H magnetic resonance spectroscopy. Magn Reson Med 49:223–232.

26. Wang R, et al. (2010) Glioblastoma stem-like cells give rise to tumour endothelium.Nature 468:829–833.

27. Ricci-Vitiani L, et al. (2010) Tumour vascularization via endothelial differentiation ofglioblastoma stem-like cells. Nature 468:824–828.

28. Wick W, Weller M, van den Bent M, Stupp R (2010) Bevacizumab and recurrentmalignant gliomas: A European perspective. J Clin Oncol 28(12):e188–e189; authorreply e190–e182.

29. Leach MO, et al.; Pharmacodynamic/Pharmacokinetic Technologies Advisory Committee,Drug Development Office, Cancer Research UK (2005) The assessment of antiangiogenicand antivascular therapies in early-stage clinical trials using magnetic resonanceimaging: Issues and recommendations. Br J Cancer 92:1599–1610.

30. Hylton N (2006) Dynamic contrast-enhanced magnetic resonance imaging as animaging biomarker. J Clin Oncol 24:3293–3298.

31. Jain RK, et al. (2009) Biomarkers of response and resistance to antiangiogenictherapy. Nat Rev Clin Oncol 6:327–338.

32. Chenevert TL, et al. (2000) Diffusion magnetic resonance imaging: An early surrogatemarker of therapeutic efficacy in brain tumors. J Natl Cancer Inst 92:2029–2036.

33. Gerstner ER, Frosch MP, Batchelor TT (2010) Diffusion magnetic resonance imagingdetects pathologically confirmed, nonenhancing tumor progression in a patient withrecurrent glioblastoma receiving bevacizumab. J Clin Oncol 28:e91–e93.

34. Folkman J (1972) Anti-angiogenesis: New concept for therapy of solid tumors. AnnSurg 175:409–416.

35. Ellis LM, Hicklin DJ (2008) Pathways mediating resistance to vascular endothelialgrowth factor-targeted therapy. Clin Cancer Res 14:6371–6375.

36. Batchelor TT, et al. (2007) AZD2171, a pan-VEGF receptor tyrosine kinase inhibitor,normalizes tumor vasculature and alleviates edema in glioblastoma patients. CancerCell 11:83–95.

37. Gatenby RA, Gillies RJ (2004) Why do cancers have high aerobic glycolysis? Nat RevCancer 4:891–899.

38. Lucio-Eterovic AK, Piao Y, de Groot JF (2009) Mediators of glioblastoma resistance andinvasionduringantivascularendothelialgrowthfactor therapy.ClinCancerRes15:4589–4599.

39. Pàez-Ribes M, et al. (2009) Antiangiogenic therapy elicits malignant progression oftumors to increased local invasion and distant metastasis. Cancer Cell 15:220–231.

40. Beckner ME, et al. (2005) Glycolytic glioma cells with active glycogen synthase aresensitive to PTEN and inhibitors of PI3K and gluconeogenesis. Lab Invest 85:1457–1470.

41. Ji H, et al. (2009) EGF-induced ERK activation promotes CK2-mediated disassociationof alpha-Catenin from beta-Catenin and transactivation of beta-Catenin. Mol Cell 36:547–559.

42. Bjerkvig R, Johansson M, Miletic H, Niclou SP (2009) Cancer stem cells andangiogenesis. Semin Cancer Biol 19:279–284.

43. Robey RB, Hay N (2009) Is Akt the “Warburg kinase”? Akt-energy metabolisminteractions and oncogenesis. Semin Cancer Biol 19:25–31.

44. Kroemer G, Pouyssegur J (2008) Tumor cell metabolism: Cancer’s Achilles’ heel. CancerCell 13:472–482.

45. Pathania D, Millard M, Neamati N (2009) Opportunities in discovery and delivery ofanticancer drugs targeting mitochondria and cancer cell metabolism. Adv Drug DelivRev 61:1250–1275.

46. Michelakis ED, Webster L, Mackey JR (2008) Dichloroacetate (DCA) as a potentialmetabolic-targeting therapy for cancer. Br J Cancer 99:989–994.

47. Michelakis ED (2010) Metabolic modulation of glioblastoma with dichloroacetate. SciTransl Med 2(31):31–34.

48. Provencher SW (2001) Automatic quantitation of localized in vivo 1H spectra withLCModel. NMR Biomed 14:260–264.

3754 | www.pnas.org/cgi/doi/10.1073/pnas.1014480108 Keunen et al.

Dow

nloa

ded

by g

uest

on

Mar

ch 2

, 202

1